Two-parameter strong laws and maximal inequalities forU-statistics

From MaRDI portal
Publication:3768086





For each Borel function f on the unit square \([0,1]^ 2\) and a sequence \(\{X_ k\}\) of independent identically distributed random variables with values in [0,1] define the sequence \[ U_{n,m}(f)=\sum^{n,m}_{i,j=1,i\neq j}f(x_ i,X_ j). \] In the paper the asymptotic behavior of \(U_{n,m}(f)\) as n,m\(\to \infty\) is studied. In particular it is proved that if \(f\in L^ 1([0,1]^ 2)\) then \[ \lim_{n,m\to \infty}(1/nm)U_{n,m}(f)=\int^{1}_{0}\int^{1}_{0}f(x,y)dxdy\quad a.s. \] if and only if \(\delta_ 0(f)<\infty\), where \[ \delta_ 0(f)=\int^{1}_{0}\int^{1}_{0}| f(x,y)| (1+\log_+\frac{\| f\|_ 1| f(x,y)|}{\| f(x,\cdot)\|_ 1\| f(\cdot,y)\|_ 1})dxdy. \] The functional \(\delta_ 0(f)\) plays an important part in these investigations. It is scalar homogeneous and \(\delta_ 0(f+g)\leq \alpha \delta_ 0(f)+\alpha \delta_ 0(g)\) for some \(\alpha >1\).











This page was built for publication: Two-parameter strong laws and maximal inequalities forU-statistics

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3768086)